memlab

While looking through trace sampling and percentile calculations in , I noticed a few boundary edge cases in (): 1. 100th percentile returns : When , evaluates to . Since , it tries to return , which evaluates to (for example, accessing index on a length-5 array instead of the maximum element at index ). 2. Empty array returns (causing in callers): For an empty array , and it returns (). When this flows into via , produces , which causes subsequent `` Happy to submit a quick PR with unit tests if this looks good to you!

While exploring the programmatic API in , I noticed a small configuration mismatch in (). In (line 156) and (line 191), is initialized with the run's config instance: However, in (line 226), is constructed without arguments: Because is omitted, defaults its internal to the global singleton . When options like a transient working directory, custom , , or custom scenario settings are passed to , the scenario synthesizers instantiated inside receive instead of the newly created instance for this run. Proposed fix: Pass to in , matching and : I'd be happy to open a PR for this if this looks good to the maintainers!

Problem When is enabled, double-counts the representative trace of every cluster: The HAC union-find labels in point every trace at the largest index in its cluster, and exactly one trace per cluster satisfies (the representative). For that one trace, , so it is inserted by the seed and pushed again by the loop — every cluster ends up containing its representative twice. This propagates to , so and are inflated by 1 per cluster, and even a single leaked object reports . Fix Initialize each cluster list empty and let the loop add every trace exactly once. Test Added , which fails before the fix (single trace → 2) and passes after (each trace counted once). All core and heap-analysis tests pass.